fleetd #621: make the context-roll notice obey requireOperatorConfirm

contextNotice hardcoded 'ask the operator' and 'Only the operator can
approve the roll', so setting leadRollover.requireOperatorConfirm to
false stopped the daemon refusing the roll but never stopped the lead
being told to ask. Thread the effective config value into
contextNotice: when true the text stays byte-identical, when false it
tells the lead to confirm on its own judgement against the three
handover-file checks instead.

LeadRollover.confirm's own enforcement is untouched — this is the
message only.

fleetd/src/main/java/dev/ltms/fleet/Fleetd.java

@@ -53,6 +53,7 @@ import dev.ltms.fleet.rest.FleetApp;
 import dev.ltms.fleet.session.GitWorktrees;
 import dev.ltms.fleet.session.MemberSession;
 import dev.ltms.fleet.session.SessionManager;
+import dev.ltms.fleet.peer.MemberRole;
 import dev.ltms.fleet.peer.PeerLauncher;
 import dev.ltms.fleet.session.SessionReaper;
 import dev.ltms.fleet.member.ClaudeCodeLauncher;
@@ -176,6 +177,12 @@ public final class Fleetd {
         // and the Fleetd-startup tests actually pin — see FleetConfig#validateAll's javadoc for
         // why a name-by-name list here would have the same defect it replaces.
         cfg.validateAll();
+        // fleetd #613: validateAll() (validateMembers() inside it) only refuses a slot that names a
+        // bad role or profile — it says nothing about a role that has NO pool or NO charter at all,
+        // because both are legitimate ("unconstrained") states, not errors. Report them here, right
+        // after validation passes, so an operator sees the gap once per restart instead of finding
+        // it later in a roster row (see reportRoleFallbackGaps' javadoc for the measured cause).
+        reportRoleFallbackGaps(cfg);
         // fleetd #469, follow-up to #464: validateAll() (and validateCharters() inside it) only
         // checks that a charter's KEY is a role wire name and its text is non-blank — it never
         // looks at what the text actually names. This is the separate check that does: it asks
@@ -570,13 +577,18 @@ public final class Fleetd {
             // (keyed by configDir+sessionId), so this costs at most one extra bounded tail read per
             // TTL window, never a shared-mutable-state hazard between the two callers.
             var leadContextGauge = new LeadContextGauge();
+            // fleetd #621: the context-high notice's own wording must track this same effective
+            // value — LeadRollover.confirm(...) already gates the roll on it (LeadRollover.java:480),
+            // and absent `leadRollover:` entirely the roll is unusable regardless (NOT_CONFIGURED),
+            // so `true` (the FleetConfig.LeadRollover default) is the safe, byte-identical fallback.
+            boolean requireOperatorConfirm = cfg.leadRollover() == null || cfg.leadRollover().requireOperatorConfirm();
             heartbeat = new LeadHeartbeatLoop(primaryRegistry, router.leadAgents(), replyInbox, sessions::roster,
                     pushLoop, heartbeatScheduler, System::nanoTime,
                     TimeUnit.SECONDS.toNanos(hb.idleAfterSeconds()), hb.backoffMs(), hb.quietNudgeCap(),
                     metrics,
                     leadContextSource(leadContextGauge, router.leadAgents(), leads,
                             leadConfigDirLookup(() -> config.get().profiles(), leaders)),
-                    Boolean.TRUE.equals(hb.contextHighNudge()));
+                    Boolean.TRUE.equals(hb.contextHighNudge()), requireOperatorConfirm);
             heartbeat.start();
         } else {
             heartbeat = null;
@@ -2175,6 +2187,58 @@ public final class Fleetd {
         }
     }
 
+    /**
+     * fleetd #613: {@code FleetConfig.candidateProfiles(MemberRole)} (FleetConfig.java:1827) and
+     * {@code CompositePeerLauncher.poolFor} (CompositePeerLauncher.java:598-603) both fall back to
+     * <em>every</em> configured profile when a role has no {@code fleet.<role>s:} pool — a
+     * deliberate "unconstrained" behaviour, kept unchanged here, that lets a config with only
+     * {@code profiles:} and no {@code fleet:} block still spawn. That fallback is silent, and on
+     * the host that opened this ticket it widened an unqualified {@code hunter} spawn to all 8
+     * configured profiles and picked {@code local} as the resolved first choice — a profile every
+     * other pool on that same config gives weight 0 to. Report it once at boot instead, naming both
+     * how many profiles the gap opens onto and the exact first choice, since the first choice (not
+     * the pool size) is what actually surprised the operator.
+     *
+     * <p>A missing {@code fleet.charters.<role>:} entry is reported separately: the member still
+     * runs, but with only the launcher's own reply charter and no role contract. Unlike the pool
+     * gap this already has a per-spawn instrument ({@code HerdrPeerLauncher.logCharterReceipt},
+     * {@code SessionManager}'s {@code charterSource} roster field) — this boot line is the same
+     * information surfaced once, up front, rather than discovered per member later.
+     *
+     * <p>Never refuses to start over either gap — both are legitimate configurations, and this is a
+     * report, not a validation. Package-private so a test can call it directly and capture the log
+     * via a {@link ch.qos.logback.core.read.ListAppender}, the same pattern {@link
+     * #reportExhaustedPatternGap} and {@link #reportMemberCredentialsGap} already use.
+     */
+    static void reportRoleFallbackGaps(FleetConfig cfg) {
+        List<String> poolGaps = new ArrayList<>();
+        List<String> charterGaps = new ArrayList<>();
+        int profileCount = cfg.profiles().size();
+        for (MemberRole role : MemberRole.values()) {
+            boolean hasPool = cfg.fleet() != null && !cfg.fleet().profilesFor(role).isEmpty();
+            if (!hasPool) {
+                String firstChoice = cfg.defaultProfileFor(role);
+                poolGaps.add(role.wireName() + " (may land on any of " + profileCount
+                        + " profile(s), first choice "
+                        + (firstChoice == null ? "none — no profiles configured" : "'" + firstChoice + "'")
+                        + ")");
+            }
+            String charter = cfg.fleet() == null ? null : cfg.fleet().charterFor(role);
+            if (charter == null || charter.isBlank()) {
+                charterGaps.add(role.wireName());
+            }
+        }
+        if (!poolGaps.isEmpty()) {
+            log.info("role fallback: no fleet.<role>s: pool for {} — an unqualified spawn of that "
+                    + "role falls back to every configured profile (deliberate; see "
+                    + "FleetConfig#candidateProfiles)", poolGaps);
+        }
+        if (!charterGaps.isEmpty()) {
+            log.info("role fallback: no fleet.charters: entry for {} — that role runs with only "
+                    + "the launcher's reply charter, no role contract", charterGaps);
+        }
+    }
+
     /**
      * fleetd #474: the one place both the startup call (right after {@code cfg.validateAll()} in
      * {@link #main}) and the reload call (wired into {@code config}'s {@code extraValidation} above,

fleetd/src/main/java/dev/ltms/fleet/config/FleetConfig.java

@@ -2240,11 +2240,11 @@ public record FleetConfig(
      * information — which profiles, and now both values, so they can fix it without reading the
      * source — without ever taking the fleet down.
      *
-     * <p>Which of the two inputs Claude Code actually follows when they disagree is intentionally
-     * <em>not</em> asserted here. {@code ClaudeCodeArguments}'s javadoc used to state the
-     * environment variable always wins; nobody had measured that, and this host's own
-     * {@code fleetd.yaml} asserts the opposite in a comment. This method only detects and reports
-     * the disagreement — see {@link dev.ltms.fleet.launch.ClaudeCodeArguments}.
+     * <p>fleetd #618 measured which of the two inputs Claude Code actually follows when they
+     * disagree: the environment variable wins, so {@code autoCompactWindow} is inert on a profile
+     * that also sets the env var. This method only detects and reports the disagreement — it does
+     * not correct it — see {@link dev.ltms.fleet.launch.ClaudeCodeArguments} for the full measured
+     * precedence.
      *
      * <p>Equal values never warn: either input then produces the same session window, so there is
      * nothing to reconcile.
@@ -2285,9 +2285,11 @@ public record FleetConfig(
         names.sort(String::compareTo);
         detail.sort(String::compareTo);
         log.warn("Claude Code profile(s) {} set disagreeing autoCompactWindow and env."
-                        + "CLAUDE_CODE_AUTO_COMPACT_WINDOW — the daemon starts anyway. Fix by "
-                        + "removing one key or setting equal values on each: {}. Which input Claude "
-                        + "Code actually follows when they disagree is not verified here.",
+                        + "CLAUDE_CODE_AUTO_COMPACT_WINDOW — the daemon starts anyway: {}. fleetd "
+                        + "#618 measured that CLAUDE_CODE_AUTO_COMPACT_WINDOW wins, so "
+                        + "autoCompactWindow is inert on these profiles. Set equal values on each "
+                        + "to resolve this — do not just delete the env var, since that LOWERS the "
+                        + "live window to autoCompactWindow's value rather than fixing anything.",
                 names, String.join(", ", detail));
     }
 

fleetd/src/main/java/dev/ltms/fleet/launch/ClaudeCodeArguments.java

@@ -15,13 +15,15 @@ public final class ClaudeCodeArguments {
      * Append the configured Claude Code auto-compaction window when the profile opts in.
      *
      * <p>This flag and the environment variable {@code CLAUDE_CODE_AUTO_COMPACT_WINDOW} can
-     * disagree. Which one Claude Code actually follows when they do is NOT verified here — this
-     * javadoc used to claim the environment variable always wins, but nobody had measured that, and
-     * this host's own {@code fleetd.yaml} asserts the opposite in a comment. So this javadoc no
-     * longer picks a side. {@link FleetConfig#load(java.nio.file.Path)} only WARNS when a Claude
-     * Code profile sets both to different values (see {@code
+     * disagree, and fleetd #618 measured which one Claude Code actually follows: the environment
+     * variable wins, ahead of this {@code --autocompact} flag, ahead of the settings file, ahead of
+     * clientdata, the experiment, and the model default. So when a profile sets both, the flag this
+     * method appends has NO effect — Claude Code reads {@code CLAUDE_CODE_AUTO_COMPACT_WINDOW}
+     * first and never consults the flag. {@link FleetConfig#load(java.nio.file.Path)} only WARNS
+     * when a Claude Code profile sets both to different values (see {@code
      * FleetConfig.warnConflictingAutoCompactWindows}) — it does not stop the daemon from starting,
-     * and a launched session may end up honouring either window.
+     * and the launched session honours the env var, not this flag. Measured against Claude Code
+     * 2.1.278 (fleetd #618) — a later version could reorder this precedence.
      */
     public static List<String> withAutoCompactWindow(List<String> argv, FleetConfig.Profile profile) {
         if (profile.autoCompactWindow() == null) {

fleetd/src/main/java/dev/ltms/fleet/lead/LeadRollover.java

@@ -231,7 +231,9 @@ public final class LeadRollover {
          * #status} could wrongly answer {@link #UNKNOWN} ("nothing was ever requested") for a roll
          * that is, in fact, actively running. This is not sticky: the deferred continuation
          * overwrites this same entry with a terminal state ({@link #ROLLED}, {@link
-         * #TURN_NEVER_SETTLED}, or {@link #CLEAR_NEVER_SETTLED}) once it finishes.
+         * #TURN_NEVER_SETTLED}, {@link #CLEAR_NEVER_SETTLED}, or {@link #FAILED}) once it finishes
+         * — including by throwing, which fleetd #615's catch in {@link #runRollover} now turns into
+         * {@link #FAILED} instead of leaving this entry stuck forever.
          */
         IN_PROGRESS,
         /**
@@ -253,6 +255,19 @@ public final class LeadRollover {
          * within {@code clearSettleSeconds} — {@code bootstrapText} was never sent.
          */
         CLEAR_NEVER_SETTLED,
+        /**
+         * fleetd #615: the deferred continuation threw a {@link RuntimeException} — most likely a
+         * {@link dev.ltms.fleet.herdr.HerdrException} out of one of the two unwrapped {@code
+         * agents.send} calls in {@link #runRollover} — and the continuation thread died with it.
+         * Before this state existed, that throw left {@link #outcomes} holding {@link #IN_PROGRESS}
+         * forever, because the production {@code continuationRunner} is a bare virtual thread with
+         * no uncaught-exception handler and nothing downstream of the throw ever ran to write a
+         * terminal outcome. {@code detail} names the exception, so a reader has something to act on
+         * — the same diagnostic style as {@link #TURN_NEVER_SETTLED} and {@link
+         * #CLEAR_NEVER_SETTLED}. The roll is dead at this point and does not retry itself; a stuck
+         * lead must {@link #open} a fresh request.
+         */
+        FAILED,
         /**
          * {@code token} names nothing this instance currently knows about: never issued by {@link
          * #open}, dropped by {@link #cancel}, or aged out of {@link #outcomes}'s bounded history.
@@ -493,8 +508,42 @@ public final class LeadRollover {
      * entirely after {@link #confirm} has returned to its caller — see this class's javadoc for the
      * four-step order. There is no result to return to by this point, so every outcome is logged
      * only.
+     *
+     * <p><strong>fleetd #615 — the whole body is wrapped in one {@code try}.</strong> The two {@code
+     * agents.send} calls below are not wrapped individually: {@code send} → {@code agentCall} →
+     * {@code herdr.call} can throw an unchecked {@link dev.ltms.fleet.herdr.HerdrException} (see
+     * {@code AgentControl.java}), and the production {@code continuationRunner} is a bare virtual
+     * thread with no uncaught-exception handler (see this class's public constructor). Before this
+     * fix, either throw killed the continuation thread silently, leaving the {@link
+     * RollState#IN_PROGRESS} entry {@link #confirm} wrote at hand-off stuck forever — {@link
+     * #status} had no way to tell a dead roll from one still genuinely running. The {@code catch}
+     * below is scoped to the method body rather than to each {@code send} call individually, so it
+     * also covers anything else added to this continuation later, not just today's two call sites —
+     * the same reasoning that put the write-a-terminal-outcome step at each of this method's other
+     * exits (see the {@link RollState#TURN_NEVER_SETTLED} and {@link RollState#CLEAR_NEVER_SETTLED}
+     * branches below) rather than inside the helpers that detect them.</p>
+     *
+     * <p>Only {@link RuntimeException} is caught, matching the local convention {@link
+     * #waitUntilAtTurnBoundary} already set around its own {@code agents.status} call — not the
+     * broader {@link Exception} or {@link Throwable}, which would also swallow something like an
+     * {@link OutOfMemoryError} this continuation has no business handling.</p>
      */
     private void runRollover(PendingRollover p, FleetConfig.LeadRollover cfg) {
+        try {
+            runRolloverUnguarded(p, cfg);
+        } catch (RuntimeException e) {
+            log.warn("lead-rollover: continuation for token={} lead={} threw {} — the roll is dead; "
+                            + "no further step in this continuation will run",
+                    p.token(), p.leadTerminal(), e.toString(), e);
+            outcomes.put(p.token(), new RollStatus(RollState.FAILED,
+                    "the roll's continuation threw " + e.toString() + " — the roll is dead and will "
+                            + "not retry itself; check the daemon log for the stack trace, then open() "
+                            + "a fresh rollover request"));
+        }
+    }
+
+    /** The actual body of {@link #runRollover}, unwrapped — see that method's javadoc for the catch. */
+    private void runRolloverUnguarded(PendingRollover p, FleetConfig.LeadRollover cfg) {
         String lead = p.leadTerminal();
         long rollStartMillis = nowMillis.getAsLong();
         TurnSettleResult turnResult = waitUntilAtTurnBoundary(lead, cfg.turnSettleSeconds());

fleetd/src/main/java/dev/ltms/fleet/msg/LeadHeartbeatLoop.java

@@ -76,6 +76,7 @@ public final class LeadHeartbeatLoop {
     private final Metrics metrics; // CB-512 pattern: nullable — no registry in unit tests
     private final LeadContextSource contextSource; // fleetd #609
     private final boolean contextHighNudge; // fleetd #609: opt-in, like the loop itself
+    private final boolean requireOperatorConfirm; // fleetd #621: mirrors leadRollover.requireOperatorConfirm
 
     /** When the current idle stretch began (nanos), or {@link #NOT_IDLE}. Single scheduler thread only. */
     private long idleSinceNanos = NOT_IDLE;
@@ -106,12 +107,32 @@ public final class LeadHeartbeatLoop {
      * fleetd #609: as above, plus the lead's own context source and whether a HIGH reading should
      * append a hand-over notice to the loop's nudge. Pass {@link LeadContextSource#none()} and
      * {@code false} to keep the pre-#609 behaviour exactly (both existing public constructors do).
+     *
+     * <p>fleetd #621: delegates to the full constructor with {@code requireOperatorConfirm=true} —
+     * the pre-#621 wording ("ask the operator ... only the operator can approve the roll") assumed
+     * the config default, so every caller of this overload keeps that text byte-identical.
      */
     public LeadHeartbeatLoop(PrimaryRegistry primaryRegistry, AgentControl agents, ReplyInbox inbox,
                              Supplier<List<MemberSession>> roster, ReplyPushLoop pushLoop,
                              ScheduledExecutorService scheduler, LongSupplier clock,
                              long idleAfterNanos, long backoffMs, int quietNudgeCap, Metrics metrics,
                              LeadContextSource contextSource, boolean contextHighNudge) {
+        this(primaryRegistry, agents, inbox, roster, pushLoop, scheduler, clock,
+                idleAfterNanos, backoffMs, quietNudgeCap, metrics, contextSource, contextHighNudge, true);
+    }
+
+    /**
+     * fleetd #621: as above, plus the daemon's effective {@code leadRollover.requireOperatorConfirm}
+     * value — threaded into {@link #contextNotice(boolean, LeadContextGauge.Reading, boolean, boolean)}
+     * so the notice's wording tracks the config the daemon actually enforces (see {@code
+     * LeadRollover.confirm}) instead of always asserting the operator gate is on.
+     */
+    public LeadHeartbeatLoop(PrimaryRegistry primaryRegistry, AgentControl agents, ReplyInbox inbox,
+                             Supplier<List<MemberSession>> roster, ReplyPushLoop pushLoop,
+                             ScheduledExecutorService scheduler, LongSupplier clock,
+                             long idleAfterNanos, long backoffMs, int quietNudgeCap, Metrics metrics,
+                             LeadContextSource contextSource, boolean contextHighNudge,
+                             boolean requireOperatorConfirm) {
         this.primaryRegistry = primaryRegistry;
         this.agents = agents;
         this.inbox = inbox;
@@ -125,6 +146,7 @@ public final class LeadHeartbeatLoop {
         this.metrics = metrics;
         this.contextSource = contextSource;
         this.contextHighNudge = contextHighNudge;
+        this.requireOperatorConfirm = requireOperatorConfirm;
     }
 
     /**
@@ -271,8 +293,13 @@ public final class LeadHeartbeatLoop {
      * does not evaluate the lead's idle state before the fleet has settled.
      */
     public void start() {
-        log.info("idle-lead heartbeat: on — nudge lead after {}s idle (recheck {}ms, quiet cap {})",
-                TimeUnit.NANOSECONDS.toSeconds(idleAfterNanos), backoffMs, quietNudgeCap);
+        // fleetd #613: contextHighNudge added alongside the three settings already here — an
+        // operator otherwise cannot tell from the boot log whether the #609 handover notice is
+        // armed, and had to load the deployed jar's config to confirm it.
+        log.info("idle-lead heartbeat: on — nudge lead after {}s idle (recheck {}ms, quiet cap {}, "
+                        + "context-high nudge {})",
+                TimeUnit.NANOSECONDS.toSeconds(idleAfterNanos), backoffMs, quietNudgeCap,
+                contextHighNudge);
         scheduler.schedule(this::tick, backoffMs, TimeUnit.MILLISECONDS);
     }
 
@@ -339,7 +366,7 @@ public final class LeadHeartbeatLoop {
         // d.contextNotified() is the value to persist once delivery is confirmed, not the value the text
         // itself should be built from. Otherwise a HIGH stretch that is still latched would never see the
         // notice at all, defeating the very check this fixes.
-        String notice = contextNotice(contextHighNudge, reading, contextNotified);
+        String notice = contextNotice(contextHighNudge, reading, contextNotified, requireOperatorConfirm);
         var lead = primaryRegistry.primaryTerminal();
         boolean sent = lead.isPresent() && trySend(lead.get(), fleet.nudgeText() + notice, notice);
         // The latch becomes true only when all three hold: decide() chose to notify, a notice was
@@ -376,8 +403,9 @@ public final class LeadHeartbeatLoop {
     /**
      * fleetd #609: the text appended to a nudge when the lead's own context is full — {@code ""}
      * whenever the notice does not apply, so callers can unconditionally append this without an extra
-     * branch. Wording stays plain (CEFR B1) and honest that only the operator approves a roll — this
-     * loop only ever prints text, it never calls {@code fleet_handover} itself.
+     * branch. Wording stays plain (CEFR B1) and honest about who actually gates the roll — see the
+     * {@code requireOperatorConfirm} overload (fleetd #621) for which check that is. This loop only
+     * ever prints text, it never calls {@code fleet_handover} itself.
      *
      * @param enabled the {@code leadHeartbeat.contextHighNudge} config flag
      * @param reading the lead's current {@link LeadContextGauge} reading
@@ -396,9 +424,33 @@ public final class LeadHeartbeatLoop {
      * closing sentence ("You will not be told again until your context reads ok.") false. {@link
      * #injectNudge} is the only caller that passes a non-default {@code alreadyNotified}.
      *
+     * <p>fleetd #621: delegates with {@code requireOperatorConfirm=true} — the pre-#621 default and the
+     * value every existing caller of this overload (including every test written before #621) already
+     * assumed, so the text this overload returns stays byte-identical.
+     *
      * @param alreadyNotified whether the lead has already been told about the current HIGH stretch
      */
     static String contextNotice(boolean enabled, LeadContextGauge.Reading reading, boolean alreadyNotified) {
+        return contextNotice(enabled, reading, alreadyNotified, true);
+    }
+
+    /**
+     * fleetd #621: as {@link #contextNotice(boolean, LeadContextGauge.Reading, boolean)}, but the closing
+     * instructions also track the daemon's effective {@code leadRollover.requireOperatorConfirm} value,
+     * instead of always asserting that only the operator can approve the roll.
+     *
+     * <p>{@code LeadRollover.confirm(...)} already honours this flag: when it is {@code false}, the daemon
+     * itself gates the roll on the three handover-file checks alone (exists, modified after the {@code
+     * open()} request, and no older than {@code maxDocAgeSeconds}) and never consults {@code
+     * operatorConfirmed}. Before this parameter existed, this notice told the lead to ask the operator
+     * regardless — so a lead that followed its own instructions asked anyway, and setting the config knob
+     * to {@code false} stopped the daemon refusing the roll without stopping the operator being
+     * interrupted. This parameter is how the text is kept honest about which gate is actually live.
+     *
+     * @param requireOperatorConfirm the effective {@code leadRollover.requireOperatorConfirm} value
+     */
+    static String contextNotice(boolean enabled, LeadContextGauge.Reading reading, boolean alreadyNotified,
+                                boolean requireOperatorConfirm) {
         if (!enabled || alreadyNotified || reading.state() != LeadContextGauge.State.HIGH) {
             return "";
         }
@@ -415,10 +467,18 @@ public final class LeadHeartbeatLoop {
             sb.append(" (").append(reading.compactions()).append(' ').append(compactionWord)
                     .append(" so far).");
         }
-        sb.append(" A fresh session would work better. To hand over: call fleet_handover(action=\"open\"), "
-                + "write the file it names, ask the operator, then call fleet_handover(action=\"confirm\", "
-                + "token, operatorConfirmed). Only the operator can approve the roll. You will not be told "
-                + "again until your context reads ok.");
+        if (requireOperatorConfirm) {
+            sb.append(" A fresh session would work better. To hand over: call fleet_handover(action=\"open\"), "
+                    + "write the file it names, ask the operator, then call fleet_handover(action=\"confirm\", "
+                    + "token, operatorConfirmed). Only the operator can approve the roll. You will not be told "
+                    + "again until your context reads ok.");
+        } else {
+            sb.append(" A fresh session would work better. To hand over: call fleet_handover(action=\"open\"), "
+                    + "write the file it names, then call fleet_handover(action=\"confirm\", token). Decide for "
+                    + "yourself when to confirm: the roll goes through if the handover file exists, was "
+                    + "changed after you opened it, and is not older than maxDocAgeSeconds. You will not be "
+                    + "told again until your context reads ok.");
+        }
         return sb.toString();
     }
 

fleetd/src/test/java/dev/ltms/fleet/RoleFallbackGapReportTest.java

@@ -0,0 +1,252 @@
+package dev.ltms.fleet;
+
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.Logger;
+import ch.qos.logback.classic.spi.ILoggingEvent;
+import ch.qos.logback.core.read.ListAppender;
+import dev.ltms.fleet.config.FleetConfig;
+import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.io.TempDir;
+import org.slf4j.LoggerFactory;
+
+import java.nio.file.Files;
+import java.nio.file.Path;
+import java.util.List;
+
+import static org.junit.jupiter.api.Assertions.assertFalse;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+/**
+ * fleetd #613: a {@code MemberRole} with no {@code fleet.<role>s:} pool falls back to
+ * <em>every</em> configured profile ({@code FleetConfig#candidateProfiles}), and one with no
+ * {@code fleet.charters.<role>:} entry runs with only the launcher's reply charter. Both are
+ * deliberate, legitimate states — neither is refused by {@code validateMembers()} — but both were
+ * silent at boot. On the host that opened this ticket, an unqualified {@code hunter} spawn silently
+ * widened to all 8 configured profiles and its resolved first choice was {@code local}, a profile
+ * every other pool on that same config gave weight 0 to.
+ *
+ * <p>{@link Fleetd#reportRoleFallbackGaps} must name every gapped role, and for a pool gap, the
+ * exact resolved first-choice profile — that number, not the pool size, is what actually surprised
+ * the operator. Mirrors {@link ExhaustedPatternGapReportTest}'s pattern: capture the real log via a
+ * {@link ListAppender} rather than asserting on the call site's source text.
+ */
+class RoleFallbackGapReportTest {
+
+    private static FleetConfig load(Path dir, String yaml) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml);
+        return FleetConfig.load(f);
+    }
+
+    /**
+     * The level this logger had before {@link #attach()} raised it, so {@link #detach} can put it
+     * back. {@code null} means "inherit from the parent" — the state this logger starts in.
+     */
+    private static Level originalLevel;
+
+    /**
+     * {@code reportRoleFallbackGaps} logs at INFO, and {@code logback-test.xml} sets
+     * {@code dev.ltms.fleet} to WARN — so INFO events are dropped by the level check before any
+     * appender sees them. Raise the level for the duration of the test, exactly like {@code
+     * GitHostShapeReportTest#attach}.
+     */
+    private static ListAppender<ILoggingEvent> attach() {
+        Logger logger = (Logger) LoggerFactory.getLogger(Fleetd.class);
+        originalLevel = logger.getLevel();
+        logger.setLevel(Level.INFO);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        logger.addAppender(appender);
+        return appender;
+    }
+
+    private static void detach(ListAppender<ILoggingEvent> appender) {
+        Logger logger = (Logger) LoggerFactory.getLogger(Fleetd.class);
+        logger.detachAppender(appender);
+        logger.setLevel(originalLevel);
+    }
+
+    private static List<String> infoMessages(ListAppender<ILoggingEvent> appender) {
+        return appender.list.stream()
+                .filter(e -> e.getLevel() == Level.INFO)
+                .map(ILoggingEvent::getFormattedMessage)
+                .toList();
+    }
+
+    /**
+     * Reproduces the shape measured in the ticket: {@code dev}, {@code reviewer} and
+     * {@code architect} each have a pool and a charter; {@code hunter} has neither. The pool-gap
+     * line must name {@code hunter}, the profile count (3), and the resolved first choice
+     * ({@code local}, the first profile in definition order) — and must not name the three healthy
+     * roles. The charter-gap line must separately name only {@code hunter}.
+     */
+    @Test
+    void hunterWithNoPoolOrCharterIsNamedWithItsResolvedFirstChoice(@TempDir Path dir) throws Exception {
+        FleetConfig cfg = load(dir, """
+                profiles:
+                  local:
+                    baseUrl: http://gx00.gw:8000
+                  sonnet:
+                    baseUrl: https://llm.ltms.dev/v1
+                  terra:
+                    baseUrl: https://llm.ltms.dev/v1
+                fleet:
+                  developers:
+                    a:
+                      profile: sonnet
+                  reviewers:
+                    b:
+                      profile: terra
+                  architects:
+                    c:
+                      profile: sonnet
+                  charters:
+                    dev: "dev charter text"
+                    reviewer: "reviewer charter text"
+                    architect: "architect charter text"
+                """);
+
+        ListAppender<ILoggingEvent> appender = attach();
+        try {
+            Fleetd.reportRoleFallbackGaps(cfg);
+        } finally {
+            detach(appender);
+        }
+
+        List<String> infos = infoMessages(appender);
+
+        String poolLine = infos.stream()
+                .filter(m -> m.contains("no fleet.<role>s: pool"))
+                .findFirst()
+                .orElseThrow(() -> new AssertionError("expected a pool-gap INFO line: " + infos));
+        assertTrue(poolLine.contains("hunter"), poolLine);
+        assertTrue(poolLine.contains("3"), "must name the profile count the gap opens onto: " + poolLine);
+        assertTrue(poolLine.contains("'local'"),
+                "must name the resolved first-choice profile, the number that actually surprised "
+                        + "the operator: " + poolLine);
+        for (String healthy : List.of("dev", "reviewer", "architect")) {
+            assertFalse(poolLine.contains(healthy),
+                    "pool-gap line must not name a role that has a pool: " + poolLine);
+        }
+
+        String charterLine = infos.stream()
+                .filter(m -> m.contains("no fleet.charters: entry"))
+                .findFirst()
+                .orElseThrow(() -> new AssertionError("expected a charter-gap INFO line: " + infos));
+        assertTrue(charterLine.contains("hunter"), charterLine);
+        for (String healthy : List.of("dev", "reviewer", "architect")) {
+            assertFalse(charterLine.contains(healthy),
+                    "charter-gap line must not name a role that has a charter: " + charterLine);
+        }
+    }
+
+    /**
+     * The resolved first choice must be genuinely computed from definition order, not hardcoded —
+     * reordering {@code profiles:} so a different entry comes first changes the reported choice.
+     */
+    @Test
+    void theResolvedFirstChoiceFollowsProfileDefinitionOrder(@TempDir Path dir) throws Exception {
+        FleetConfig cfg = load(dir, """
+                profiles:
+                  sonnet:
+                    baseUrl: https://llm.ltms.dev/v1
+                  local:
+                    baseUrl: http://gx00.gw:8000
+                fleet:
+                  developers:
+                    a:
+                      profile: sonnet
+                  charters:
+                    dev: "dev charter text"
+                """);
+
+        ListAppender<ILoggingEvent> appender = attach();
+        try {
+            Fleetd.reportRoleFallbackGaps(cfg);
+        } finally {
+            detach(appender);
+        }
+
+        String poolLine = infoMessages(appender).stream()
+                .filter(m -> m.contains("no fleet.<role>s: pool"))
+                .findFirst()
+                .orElseThrow();
+        // hunter, reviewer and architect all lack a pool here; each falls back to the full 2-profile
+        // set and the first-choice is 'sonnet' because it is first in profiles: definition order.
+        assertTrue(poolLine.contains("'sonnet'"), poolLine);
+        assertFalse(poolLine.contains("'local'"), poolLine);
+    }
+
+    /** A config with a pool and a charter for every role produces no role-fallback log at all. */
+    @Test
+    void everyRoleWithAPoolAndACharterProducesNoLogAtAll(@TempDir Path dir) throws Exception {
+        FleetConfig cfg = load(dir, """
+                profiles:
+                  sonnet:
+                    baseUrl: https://llm.ltms.dev/v1
+                fleet:
+                  developers:
+                    a:
+                      profile: sonnet
+                  reviewers:
+                    b:
+                      profile: sonnet
+                  hunters:
+                    c:
+                      profile: sonnet
+                  architects:
+                    d:
+                      profile: sonnet
+                  charters:
+                    dev: "dev charter text"
+                    reviewer: "reviewer charter text"
+                    hunter: "hunter charter text"
+                    architect: "architect charter text"
+                """);
+
+        ListAppender<ILoggingEvent> appender = attach();
+        try {
+            Fleetd.reportRoleFallbackGaps(cfg);
+        } finally {
+            detach(appender);
+        }
+
+        assertTrue(appender.list.isEmpty(),
+                "a config with no gaps must not print a per-role block: " + infoMessages(appender));
+    }
+
+    /**
+     * A config with only {@code profiles:} and no {@code fleet:} block at all must still be
+     * reported (every role is gapped, both pool and charter) rather than throwing — this is the
+     * exact shape {@code candidateProfiles}' fallback exists to keep starting.
+     */
+    @Test
+    void aConfigWithNoFleetBlockAtAllReportsEveryRoleGapped(@TempDir Path dir) throws Exception {
+        FleetConfig cfg = load(dir, """
+                profiles:
+                  sonnet:
+                    baseUrl: https://llm.ltms.dev/v1
+                """);
+
+        ListAppender<ILoggingEvent> appender = attach();
+        try {
+            Fleetd.reportRoleFallbackGaps(cfg);
+        } finally {
+            detach(appender);
+        }
+
+        List<String> infos = infoMessages(appender);
+        String poolLine = infos.stream()
+                .filter(m -> m.contains("no fleet.<role>s: pool"))
+                .findFirst()
+                .orElseThrow(() -> new AssertionError("expected a pool-gap INFO line: " + infos));
+        String charterLine = infos.stream()
+                .filter(m -> m.contains("no fleet.charters: entry"))
+                .findFirst()
+                .orElseThrow(() -> new AssertionError("expected a charter-gap INFO line: " + infos));
+        for (String role : List.of("dev", "hunter", "reviewer", "architect")) {
+            assertTrue(poolLine.contains(role), poolLine);
+            assertTrue(charterLine.contains(role), charterLine);
+        }
+    }
+}

fleetd/src/test/java/dev/ltms/fleet/lead/LeadRolloverTest.java

@@ -1359,4 +1359,92 @@ class LeadRolloverTest {
                     + "it were the measured wait duration: " + message);
         }
     }
+
+    // ---- fleetd #615: a HerdrException out of either unwrapped agents.send call must leave a ----
+    // ---- TERMINAL FAILED outcome, never a stuck IN_PROGRESS ---------------------------------------
+
+    @Test
+    @DisplayName("[fleetd #615 — 1] send() throwing on the /clear call leaves status(token) "
+            + "reporting FAILED, not stuck at IN_PROGRESS")
+    void sendThrowingOnClearLeavesStatusReportingFailed() throws IOException {
+        FakeHerdr fake = new FakeHerdr(); // default idle — the turn-settle wait passes immediately
+        HerdrClient throwsOnClear = new HerdrClient() {
+            @Override
+            public JsonNode call(String method, Object params) throws HerdrException {
+                if ("agent.prompt".equals(method) && String.valueOf(params).contains("/clear")) {
+                    throw new HerdrException("simulated herdr transport failure sending /clear");
+                }
+                return fake.call(method, params);
+            }
+
+            @Override
+            public void close() {
+                fake.close();
+            }
+        };
+        Path handover = writeHandover("handover contents");
+        AtomicLong clock = new AtomicLong(1_000);
+        LeadRollover rollover = newRollover(throwsOnClear, cfg(handover.toString()), fixedClock(clock));
+
+        LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
+        LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
+
+        assertTrue(decision.accepted(), "every synchronous gate passes; the throw happens only "
+                + "inside the deferred continuation, which this test's synchronous runner has "
+                + "already run to completion by the time confirm() returns");
+
+        LeadRollover.RollStatus status = rollover.status(pending.token());
+        assertEquals(LeadRollover.RollState.FAILED, status.state(),
+                "a HerdrException out of the /clear send must leave a TERMINAL FAILED outcome — "
+                        + "before fleetd #615's fix, the continuation thread died silently and "
+                        + "status() was stuck reporting the IN_PROGRESS confirm() wrote at hand-off, "
+                        + "forever: got " + status.state() + " / " + status.detail());
+        assertNotEquals(LeadRollover.RollState.IN_PROGRESS, status.state());
+        assertTrue(status.detail().contains("HerdrException"), "the detail must name the exception "
+                + "so an operator reading status() has something to act on: " + status.detail());
+    }
+
+    @Test
+    @DisplayName("[fleetd #615 — 2] send() throwing on the bootstrap-text call (after /clear "
+            + "succeeded and the pane settled) also leaves status(token) reporting FAILED — a "
+            + "DIFFERENT exit from the /clear-throw case above")
+    void sendThrowingOnBootstrapTextLeavesStatusReportingFailed() throws IOException {
+        FakeHerdr fake = new FakeHerdr(); // default idle throughout — both settle waits pass promptly
+        HerdrClient throwsOnBootstrapText = new HerdrClient() {
+            @Override
+            public JsonNode call(String method, Object params) throws HerdrException {
+                if ("agent.prompt".equals(method) && String.valueOf(params).contains("read the handover file")) {
+                    throw new HerdrException("simulated herdr transport failure sending bootstrapText");
+                }
+                return fake.call(method, params);
+            }
+
+            @Override
+            public void close() {
+                fake.close();
+            }
+        };
+        Path handover = writeHandover("handover contents");
+        AtomicLong clock = new AtomicLong(1_000);
+        LeadRollover rollover = newRollover(throwsOnBootstrapText, cfg(handover.toString()), fixedClock(clock));
+
+        LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
+        LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
+
+        assertTrue(decision.accepted(), "every synchronous gate passes; the throw happens only "
+                + "inside the deferred continuation, which this test's synchronous runner has "
+                + "already run to completion by the time confirm() returns");
+        assertEquals(1, promptCallCount(fake), "sanity: /clear was sent and settled — only the "
+                + "SECOND agent.prompt call (bootstrapText) threw");
+
+        LeadRollover.RollStatus status = rollover.status(pending.token());
+        assertEquals(LeadRollover.RollState.FAILED, status.state(),
+                "a HerdrException out of the bootstrapText send — a DIFFERENT exit from the /clear "
+                        + "throw, reached only after /clear already succeeded and the pane already "
+                        + "settled — must also leave a TERMINAL FAILED outcome, not a stuck "
+                        + "IN_PROGRESS: got " + status.state() + " / " + status.detail());
+        assertNotEquals(LeadRollover.RollState.IN_PROGRESS, status.state());
+        assertTrue(status.detail().contains("HerdrException"), "the detail must name the exception "
+                + "so an operator reading status() has something to act on: " + status.detail());
+    }
 }

fleetd/src/test/java/dev/ltms/fleet/msg/LeadHeartbeatLoopTest.java

@@ -1,5 +1,9 @@
 package dev.ltms.fleet.msg;
 
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.Logger;
+import ch.qos.logback.classic.spi.ILoggingEvent;
+import ch.qos.logback.core.read.ListAppender;
 import com.fasterxml.jackson.databind.JsonNode;
 import com.fasterxml.jackson.databind.ObjectMapper;
 import dev.ltms.fleet.herdr.AgentControl;
@@ -12,6 +16,7 @@ import dev.ltms.fleet.session.MemberSession;
 import org.junit.jupiter.api.AfterEach;
 import org.junit.jupiter.api.BeforeEach;
 import org.junit.jupiter.api.Test;
+import org.slf4j.LoggerFactory;
 
 import java.util.ArrayList;
 import java.util.List;
@@ -107,6 +112,58 @@ class LeadHeartbeatLoopTest {
                 "never inject into a state the loop cannot read");
     }
 
+    // ── fleetd #613: the boot line names all four heartbeat settings ──────────────────────────
+
+    /**
+     * fleetd #613: {@link LeadHeartbeatLoop#start()}'s boot line named only 3 of the 4 constructor
+     * settings — {@code contextHighNudge} (fleetd #609) was missing, so an operator could not tell
+     * from the log whether the handover notice was armed. Captures the real log via a
+     * {@link ListAppender}, raising the logger's level past {@code logback-test.xml}'s
+     * {@code dev.ltms.fleet -> WARN} override for the duration of the call — the same seam {@code
+     * GitHostShapeReportTest#attach} uses for its own INFO-level boot line.
+     */
+    private static String heartbeatBootLine(boolean contextHighNudge, ScheduledExecutorService scheduler) {
+        Logger logger = (Logger) LoggerFactory.getLogger(LeadHeartbeatLoop.class);
+        Level originalLevel = logger.getLevel();
+        logger.setLevel(Level.INFO);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        logger.addAppender(appender);
+        try {
+            LeadHeartbeatLoop l = new LeadHeartbeatLoop(
+                    new PrimaryRegistry("term_lead"), null /*agents*/, null /*inbox*/, List::of,
+                    null /*pushLoop*/, scheduler, () -> 0L, IDLE_AFTER_NANOS, 1_000L, 3, null,
+                    LeadHeartbeatLoop.LeadContextSource.none(), contextHighNudge);
+            l.start();
+        } finally {
+            logger.detachAppender(appender);
+            logger.setLevel(originalLevel);
+        }
+        return appender.list.stream()
+                .filter(e -> e.getLevel() == Level.INFO)
+                .map(ILoggingEvent::getFormattedMessage)
+                .filter(m -> m.startsWith("idle-lead heartbeat:"))
+                .findFirst()
+                .orElseThrow(() -> new AssertionError("expected the heartbeat boot line to be logged"));
+    }
+
+    @Test
+    void theBootLineNamesContextHighNudgeWhenArmed() {
+        String line = heartbeatBootLine(true, scheduler);
+        assertTrue(line.contains("300s idle"), line);
+        assertTrue(line.contains("recheck 1000ms"), line);
+        assertTrue(line.contains("quiet cap 3"), line);
+        assertTrue(line.contains("context-high nudge true"),
+                "the boot line must name the 4th setting, contextHighNudge, alongside the other "
+                        + "three: " + line);
+    }
+
+    @Test
+    void theBootLineNamesContextHighNudgeWhenOff() {
+        String line = heartbeatBootLine(false, scheduler);
+        assertTrue(line.contains("context-high nudge false"), line);
+    }
+
     // ── (b) an idle lead within the quiet period is not yet injected ───────────────────────────
 
     @Test
@@ -419,6 +476,26 @@ class LeadHeartbeatLoopTest {
         assertTrue(notice.contains("2 compactions"), notice);
     }
 
+    // ── fleetd #621: the notice must track the effective requireOperatorConfirm value ─────────────
+
+    @Test
+    void contextNoticeKeepsAskingTheOperatorWhenRequireOperatorConfirmIsTrue() {
+        var reading = new LeadContextGauge.Reading(LeadContextGauge.State.HIGH, 260_771L, 1);
+        String notice = LeadHeartbeatLoop.contextNotice(true, reading, false, true);
+        assertTrue(notice.contains("ask the operator"), notice);
+        assertTrue(notice.contains("Only the operator can approve the roll"), notice);
+    }
+
+    @Test
+    void contextNoticeDropsTheOperatorAskWhenRequireOperatorConfirmIsFalse() {
+        var reading = new LeadContextGauge.Reading(LeadContextGauge.State.HIGH, 260_771L, 1);
+        String notice = LeadHeartbeatLoop.contextNotice(true, reading, false, false);
+        assertFalse(notice.contains("ask the operator"), notice);
+        assertFalse(notice.contains("Only the operator can approve the roll"), notice);
+        assertTrue(notice.contains("fleet_handover"), notice);
+        assertTrue(notice.contains("maxDocAgeSeconds"), notice);
+    }
+
     // ── fleetd #609 review: the latch must mean "the notice reached the pane" ────────────────────
     //
     // These four drive LeadHeartbeatLoop.tick() directly (package-private, same reasoning as